Modern applications have workloads with diverse resource requirements. ML/AI workloads require expensive GPU-equipped Workers with specific CUDA libraries, video processing needs specialized encoding hardware, and data analytics may require high-memory instances. In these cases, running all Activities on the same Worker type is cost-prohibitive and inefficient.
Without separate Task Queues for different resource types, these scenarios can result in wasted resources, inefficient usage, and conflicts between dependencies.
The Temporal Validated Pattern “Route specialized workloads” uses separate Task Queues to route Activities based on their resource requirements. You can create dedicated Worker pools for GPU-intensive ML workloads, standard CPU workloads, high-memory analytics, or specialized hardware.